Dashboard
Cases
| Patient | Age / Sex | Model call | Status | Imported |
|---|---|---|---|---|
| Loading… | ||||
Try a sample ECG · جرّب تخطيط عيّنة
Three de-identified 12-lead samples, analysed with one click, no file needed.
Analyze an ECG
Do not upload an ECG that carries a patient's name or details. Uploaded ECGs are deleted automatically after 24 hours.
لا ترفع تخطيط فيه اسم مريض أو بياناته. التخطيطات المرفوعة بتنمسح لحالها بعد 24 ساعة.
12-lead + saliency
Superclass probabilities
Clinician review
Blinded reading
No case loaded.
About QALB-ECG
A deep-learning system that screens a 12-lead ECG for 5 diagnostic superclasses (NORM, MI, STTC, CD, HYP) and highlights, via Grad-CAM, where it focused.
Intended use: research and clinician decision-support only. It does NOT make autonomous diagnoses; a licensed clinician confirms or overrides every result.
⚠️ Limitations & validation (read this)
- Not a cleared medical device — no regulatory clearance, no prospective trial, reader study not yet run.
- Validated only on public PTB-XL data (held-out macro-AUROC 0.919, ECE 0.009).
- Fails on other populations: on external (US) data MI detection collapses (AUROC 0.93→0.63) and HYP catches only ~18%. No MENA/local validation exists.
- Misses cases even internally: ~28% of MI and ~44% of HYP at the shown operating point. A screening-negative result never rules these out.
- ECG-age & predicted-sex are experimental (ECG-age ±8.2 yr) — not validated endpoints.
- The model abstains ("out-of-distribution") when an input is unlike its training data.